Semantic Embeddings for Recognition and Retrieval
The success of deep learning is, to an extent, powered by the availability of large-scale annotated dataset. However, for certain tasks, collection and annotation at scale is either expensive or unattainable. This effect is exacerbated by the typical “long tail distribution”, where small fraction of categories of interest account for large fraction of data. I will discuss a class of semantic manifold embedding approaches that are designed to perform well under such conditions. Conveniently, models of this forms can parsimoniously support variety of tasks (e.g., annotation, retrieval) and can leverage data in related domains (e.g., text) to enhance recognition of entities (e.g., images) in the target domain.
Leonid Sigal is a Senior Research Scientist at Disney Research, Pittsburgh and an Adjunct Faculty of Carnegie Mellon University. He completed his Ph.D. at Brown University (2008); received B.Sc. degrees in Computer Science and Mathematics from Boston University (1999), M.A. from Boston University (1999), and M.S. from Brown University (2003). Leonid's research interests lie in the areas of computer vision, machine learning and computer graphics. He has published more than 70 papers in top venues and journals in these fields. His current research spans action recognition, object detection and categorization; transfer, structured and representation learning.